|
Ph.D. Defense Thursday, September 21, 2006 Bourns Hall – A275 10:00AM Title: Blind Source Separation and Identification and Wireless Relays Abstract: This thesis examines a number of issues in multiantenna multiuser wireless communications. Currently as a hot topic, wireless relays are important for wireless ad hoc communication networks. Deploying a relay between a source and a destination can reduce the required transmitted power from the source, and hence reduce the interference to other neighboring nodes. A relay may also be necessary when there is strong shadowing between the source and the destination. Relays can be classified as regenerative and non-regenerative or as half-duplex and full-duplex. In this thesis, we first show that for a half-duplex non-regenerative MIMO relay system, the source covariance matrix and the relay matrix can be optimized jointly to maximize the source-destination capacity. We also propose several simple full-duplex non-regenerative schemes. We evaluate the capacities of these full-duplex schemes with the optimal capacity bounds shown in the existing work. The problem of blind channel identification and blind source separation is important in single-carrier wireless communication, speech enhancement and other applications. In this thesis, we examine a two-step maximum likelihood (TSML) solution to blind identification of SIMO auto-regressive (AR) channels, which is a useful alternative to a previously developed TSML algorithm for moving-average (MA) channels. It is shown that the AR-TSML algorithm is shown to be more robust than the MA-TSML algorithm if the channel impulse responses have long tails. We then evaluate several fast algorithms for SIMO blind channel identification, including fast TSML and approximate ML (AML). The performance and complexity of these algorithms are compared extensively. Also the problem of MIMO-FIR channel identification is investigated in this thesis. We propose ICA-BIDS algorithm which combines the benefits of the second-order statistics (SOS) based algorithm and independent component analysis (ICA) based blind source separation algorithm. The new proposed ICA-BIDS algorithm relieves the constraints imposed on signal statistics in BIDS method and also can estimate the signals and channels up to permutation and scaling ambiguities. Instead, in existing ICA algorithms, the signals can be estimated only up to filtering ambiguities. Finally, the thesis examines the design of the OFDM receivers, which is robust to the inaccurate estimate of frequency offset. Both zero-forcing and minimum mean squared error criterions are considered. |
| < Prev | Next > |
|---|